diff --git a/CMakeLists.txt b/CMakeLists.txt
index a6913bd8f..fe8954708 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -57,6 +57,12 @@ endif()
 
 # Python Support
 option(PythonSupport "Build with Python support" OFF)
+# H5 support
+option(HDF5Support "Build with HDF5 support" OFF)
+# MPI support
+option(MPISupport "Build with MPI support" OFF)
+# OpenMP support
+option(OpenMPSupport "Build with OpenMP support" OFF)
 
 # Documentation
 option(Build_Documentation "Build documentation" OFF)
diff --git a/benchmarks/CMakeLists.txt b/benchmarks/CMakeLists.txt
index 00728b352..dcf10c852 100644
--- a/benchmarks/CMakeLists.txt
+++ b/benchmarks/CMakeLists.txt
@@ -7,6 +7,9 @@ foreach(bench ${all_benchs})
   add_executable(${bench_name} ${bench})
   target_link_libraries(${bench_name} ${PROJECT_NAME}::${PROJECT_NAME}_c ${PROJECT_NAME}_warnings benchmark_main)
   set_property(TARGET ${bench_name} PROPERTY RUNTIME_OUTPUT_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/${bench_dir})
+  if (CudaSupport)
+    target_compile_definitions(${bench_name} PRIVATE -DNDA_CUDA_SUPPORT)
+  endif()
   #add_bench(NAME ${bench_name} COMMAND ${bench_name} WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/${bench_dir})
   # Run clang-tidy if found
   if(CLANG_TIDY_EXECUTABLE)
diff --git a/benchmarks/blas.cpp b/benchmarks/blas.cpp
index 2d1db358b..fd17ce450 100644
--- a/benchmarks/blas.cpp
+++ b/benchmarks/blas.cpp
@@ -22,7 +22,9 @@ static void DOT(benchmark::State &state) {
   state.counters["bytesize"] = double(NBytes);
 }
 BENCHMARK_TEMPLATE(DOT, nda::vector<value_t>)->RangeMultiplier(2)->Range(Nmin, Nmax)->Unit(benchmark::kMicrosecond);   // NOLINT
+#ifdef NDA_CUDA_SUPPORT
 BENCHMARK_TEMPLATE(DOT, nda::cuvector<value_t>)->RangeMultiplier(2)->Range(Nmin, Nmax)->Unit(benchmark::kMicrosecond); // NOLINT
+#endif
 
 template <typename Matrix>
 static void GEMM(benchmark::State &state) {
@@ -36,7 +38,9 @@ static void GEMM(benchmark::State &state) {
   state.counters["bytesize"] = double(NBytes);
 }
 BENCHMARK_TEMPLATE(GEMM, nda::matrix<value_t>)->RangeMultiplier(2)->Range(Nmin, Nmax)->Unit(benchmark::kMicrosecond);   // NOLINT
+#ifdef NDA_CUDA_SUPPORT
 BENCHMARK_TEMPLATE(GEMM, nda::cumatrix<value_t>)->RangeMultiplier(2)->Range(Nmin, Nmax)->Unit(benchmark::kMicrosecond); // NOLINT
+#endif
 
 template <typename Vector, typename Matrix>
 static void GER(benchmark::State &state) {
@@ -50,7 +54,9 @@ static void GER(benchmark::State &state) {
   state.counters["bytesize"] = double(NBytes);
 }
 BENCHMARK_TEMPLATE(GER, nda::vector<value_t>, nda::matrix<value_t>)->RangeMultiplier(2)->Range(Nmin, Nmax)->Unit(benchmark::kMicrosecond); // NOLINT
+#ifdef NDA_CUDA_SUPPORT
 BENCHMARK_TEMPLATE(GER, nda::cuvector<value_t>, nda::cumatrix<value_t>)
    ->RangeMultiplier(2)
    ->Range(Nmin, Nmax)
    ->Unit(benchmark::kMicrosecond); // NOLINT
+#endif
diff --git a/benchmarks/copy.cpp b/benchmarks/copy.cpp
index 8848d4471..856c4d40f 100644
--- a/benchmarks/copy.cpp
+++ b/benchmarks/copy.cpp
@@ -10,8 +10,10 @@ using value_t = double;
 template <size_t Rank>
 using array_t = nda::array<value_t, Rank>;
 
+#ifdef NDA_CUDA_SUPPORT
 template <size_t Rank>
 using device_array_t = nda::cuarray<value_t, Rank>;
+#endif
 
 const long KBmin = 8;
 const long KBmax = 1 << 15;
@@ -27,8 +29,9 @@ static void Copy(benchmark::State &state) {
   state.counters["processed"] = double(NBytes);
 }
 BENCHMARK_TEMPLATE(Copy, array_t<1>)->RangeMultiplier(8)->Range(KBmin, KBmax);        // NOLINT
+#ifdef NDA_CUDA_SUPPORT
 BENCHMARK_TEMPLATE(Copy, device_array_t<1>)->RangeMultiplier(8)->Range(KBmin, KBmax); // NOLINT
-
+#endif
 template <typename Array>
 static void Copy1DStrided(benchmark::State &state) {
   long NBytes = state.range(0) * 1024;
@@ -44,7 +47,9 @@ static void Copy1DStrided(benchmark::State &state) {
   state.counters["step"]      = double(step);
 }
 BENCHMARK_TEMPLATE(Copy1DStrided, array_t<1>)->RangeMultiplier(8)->Range(KBmin, KBmax);        // NOLINT
+#ifdef NDA_CUDA_SUPPORT
 BENCHMARK_TEMPLATE(Copy1DStrided, device_array_t<1>)->RangeMultiplier(8)->Range(KBmin, KBmax); // NOLINT
+#endif
 
 template <typename DstArray, typename SrcArray>
 static void CopyBlockStrided(benchmark::State &state) {
@@ -63,6 +68,8 @@ static void CopyBlockStrided(benchmark::State &state) {
   state.counters["n_blocks"]  = double(n_blocks);
 }
 BENCHMARK_TEMPLATE(CopyBlockStrided, array_t<2>, array_t<2>)->RangeMultiplier(8)->Range(KBmin, KBmax);               // NOLINT
+#ifdef NDA_CUDA_SUPPORT
 BENCHMARK_TEMPLATE(CopyBlockStrided, device_array_t<2>, device_array_t<2>)->RangeMultiplier(8)->Range(KBmin, KBmax); // NOLINT
 BENCHMARK_TEMPLATE(CopyBlockStrided, array_t<2>, device_array_t<2>)->RangeMultiplier(8)->Range(KBmin, KBmax);        // NOLINT
 BENCHMARK_TEMPLATE(CopyBlockStrided, device_array_t<2>, array_t<2>)->RangeMultiplier(8)->Range(KBmin, KBmax);        // NOLINT
+#endif
diff --git a/benchmarks/gemm_batch.cpp b/benchmarks/gemm_batch.cpp
index ed2aef1b3..f58b5047c 100644
--- a/benchmarks/gemm_batch.cpp
+++ b/benchmarks/gemm_batch.cpp
@@ -27,7 +27,9 @@ static void GEMM_BATCH(benchmark::State &state) {
   state.counters["bytesize"]   = double(NBytes);
 }
 BENCHMARK_TEMPLATE(GEMM_BATCH, nda::matrix<value_t>)->RangeMultiplier(2)->Range(Nmin, Nmax)->Unit(benchmark::kMicrosecond);   // NOLINT
+#ifdef NDA_CUDA_SUPPORT
 BENCHMARK_TEMPLATE(GEMM_BATCH, nda::cumatrix<value_t>)->RangeMultiplier(2)->Range(Nmin, Nmax)->Unit(benchmark::kMicrosecond); // NOLINT
+#endif
 
 template <typename M>
 static void GEMM_VBATCH(benchmark::State &state) {
@@ -45,7 +47,9 @@ static void GEMM_VBATCH(benchmark::State &state) {
   state.counters["bytesize"]   = double(NBytes);
 }
 BENCHMARK_TEMPLATE(GEMM_VBATCH, nda::matrix<value_t>)->RangeMultiplier(2)->Range(Nmin, Nmax)->Unit(benchmark::kMicrosecond);   // NOLINT
+#ifdef NDA_CUDA_SUPPORT
 BENCHMARK_TEMPLATE(GEMM_VBATCH, nda::cumatrix<value_t>)->RangeMultiplier(2)->Range(Nmin, Nmax)->Unit(benchmark::kMicrosecond); // NOLINT
+#endif
 
 template <typename T>
 static void GEMM_BATCH_STRIDED(benchmark::State &state) {
@@ -63,4 +67,6 @@ static void GEMM_BATCH_STRIDED(benchmark::State &state) {
   state.counters["bytesize"]   = double(NBytes);
 }
 BENCHMARK_TEMPLATE(GEMM_BATCH_STRIDED, nda::array<value_t, 3>)->RangeMultiplier(2)->Range(Nmin, Nmax)->Unit(benchmark::kMicrosecond);   // NOLINT
+#ifdef NDA_CUDA_SUPPORT
 BENCHMARK_TEMPLATE(GEMM_BATCH_STRIDED, nda::cuarray<value_t, 3>)->RangeMultiplier(2)->Range(Nmin, Nmax)->Unit(benchmark::kMicrosecond); // NOLINT
+#endif
diff --git a/c++/nda/CMakeLists.txt b/c++/nda/CMakeLists.txt
index 4644eeea1..30cc6391b 100644
--- a/c++/nda/CMakeLists.txt
+++ b/c++/nda/CMakeLists.txt
@@ -30,17 +30,30 @@ install(DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR} DESTINATION include FILES_MATCHING
 configure_file(version.hpp.in version.hpp @ONLY)
 install(FILES ${CMAKE_CURRENT_BINARY_DIR}/version.hpp DESTINATION include/${PROJECT_NAME})
 
+
+# ========= Itertools ==========
+target_link_libraries(${PROJECT_NAME}_c PUBLIC itertools::itertools_c)
+
 # ========= Additional Depdencies ==========
 
 # Link against HDF5 C++ Interface
-target_link_libraries(${PROJECT_NAME}_c PUBLIC h5::h5_c)
+if(HDF5Support)
+  message(STATUS "-------- HDF5 detection -------------")
+  target_link_libraries(${PROJECT_NAME}_c PUBLIC h5::h5_c)
+endif()
 
 # Link against MPI C++ Interface
-target_link_libraries(${PROJECT_NAME}_c PUBLIC mpi::mpi_c)
+if (MPISupport)
+  message(STATUS "-------- MPI detection -------------")
+  add_subdirectory(mpi)
+endif()
 
 # OpenMP
-find_package(OpenMP REQUIRED COMPONENTS CXX)
-target_link_libraries(${PROJECT_NAME}_c PUBLIC OpenMP::OpenMP_CXX)
+if (OpenMPSupport)
+    message(STATUS "-------- OpenMP detection -------------")
+    find_package(OpenMP REQUIRED COMPONENTS CXX)
+    target_link_libraries(${PROJECT_NAME}_c PUBLIC OpenMP::OpenMP_CXX)
+endif ()
 
 # ========= Blas / Lapack ==========
 
diff --git a/c++/nda/h5.hpp b/c++/nda/h5.hpp
index 9225f1f8c..88833d281 100644
--- a/c++/nda/h5.hpp
+++ b/c++/nda/h5.hpp
@@ -10,12 +10,12 @@
 
 #pragma once
 
-#include "./concepts.hpp"
-#include "./declarations.hpp"
-#include "./exceptions.hpp"
-#include "./layout/for_each.hpp"
-#include "./layout/range.hpp"
-#include "./traits.hpp"
+#include "concepts.hpp"
+#include "declarations.hpp"
+#include "exceptions.hpp"
+#include "layout/for_each.hpp"
+#include "layout/range.hpp"
+#include "traits.hpp"
 
 #include <h5/h5.hpp>
 
diff --git a/c++/nda/mpi.hpp b/c++/nda/mpi.hpp
index 8fe4bc6c8..9f52b8dc4 100644
--- a/c++/nda/mpi.hpp
+++ b/c++/nda/mpi.hpp
@@ -10,8 +10,8 @@
 
 #pragma once
 
-#include "./mpi/broadcast.hpp"
-#include "./mpi/gather.hpp"
-#include "./mpi/reduce.hpp"
-#include "./mpi/scatter.hpp"
-#include "./mpi/utils.hpp"
+#include "mpi/broadcast.hpp"
+#include "mpi/gather.hpp"
+#include "mpi/reduce.hpp"
+#include "mpi/scatter.hpp"
+#include "mpi/utils.hpp"
diff --git a/c++/nda/mpi/CMakeLists.txt b/c++/nda/mpi/CMakeLists.txt
new file mode 100644
index 000000000..cdeaf3abb
--- /dev/null
+++ b/c++/nda/mpi/CMakeLists.txt
@@ -0,0 +1,6 @@
+target_link_libraries(${PROJECT_NAME}_c PUBLIC mpi::mpi_c)
+
+target_compile_definitions(${PROJECT_NAME}_c PRIVATE
+        $<BUILD_INTERFACE:MPI_SUPPORT>
+        $<INSTALL_INTERFACE:MPI_SUPPORT>
+)
\ No newline at end of file
diff --git a/c++/nda/sym_grp.hpp b/c++/nda/sym_grp.hpp
index 225c4527e..051d20d01 100644
--- a/c++/nda/sym_grp.hpp
+++ b/c++/nda/sym_grp.hpp
@@ -11,10 +11,12 @@
 #pragma once
 
 #include "./nda.hpp"
-#include "./mpi.hpp"
+#ifdef MPI_SUPPORT
+#include "mpi.hpp"
+#endif
 
+#include <itertools/itertools.hpp>
 #include <itertools/omp_chunk.hpp>
-#include <mpi/mpi.hpp>
 
 #include <array>
 #include <concepts>
@@ -182,7 +184,7 @@ namespace nda {
       if (parallel) {
         // reset input array to allow for mpi reduction
         a() = 0.0;
-
+#ifdef MPI_SUPPORT
 #pragma omp parallel
         for (auto const &sym_class : itertools::omp_chunk(mpi::chunk(sym_classes))) {
           auto idx           = a.indexmap().to_idx(sym_class[0].first);
@@ -190,9 +192,25 @@ namespace nda {
           std::apply(a, idx) = ref_val;
           for (auto const &[lin_idx, op] : sym_class) { std::apply(a, a.indexmap().to_idx(lin_idx)) = op(ref_val); }
         }
-
         // distribute data among all ranks
         a = mpi::all_reduce(a);
+#elifdef _OPENMP
+#pragma omp parallel
+        for (auto const &sym_class : itertools::omp_chunk(sym_classes)) {
+          auto idx           = a.indexmap().to_idx(sym_class[0].first);
+          auto ref_val       = init_func(idx);
+          std::apply(a, idx) = ref_val;
+          for (auto const &[lin_idx, op] : sym_class) { std::apply(a, a.indexmap().to_idx(lin_idx)) = op(ref_val); }
+        }
+#else
+        for (auto const &sym_class : sym_classes) {
+          auto idx           = a.indexmap().to_idx(sym_class[0].first);
+          auto ref_val       = init_func(idx);
+          std::apply(a, idx) = ref_val;
+          for (auto const &[lin_idx, op] : sym_class) { std::apply(a, a.indexmap().to_idx(lin_idx)) = op(ref_val); }
+        }
+#endif
+
       } else {
         for (auto const &sym_class : sym_classes) {
           auto idx           = a.indexmap().to_idx(sym_class[0].first);
diff --git a/deps/CMakeLists.txt b/deps/CMakeLists.txt
index 8101214c8..a5ef4fbcc 100644
--- a/deps/CMakeLists.txt
+++ b/deps/CMakeLists.txt
@@ -84,19 +84,21 @@ external_dependency(itertools
 )
 
 # -- h5 --
-external_dependency(h5
-  GIT_REPO https://github.com/TRIQS/h5
-  VERSION 1.3
-  GIT_TAG unstable
-)
-
+if (HDF5Support)
+    external_dependency(h5
+      GIT_REPO https://github.com/TRIQS/h5
+      VERSION 1.3
+      GIT_TAG unstable
+    )
+endif()
 # -- MPI --
+if (MPISupport)
 external_dependency(mpi
   GIT_REPO https://github.com/TRIQS/mpi
   VERSION 1.3
   GIT_TAG unstable
 )
-
+endif ()
 ## Pybind 11
 #find_package(Python)
 #add_subdirectory(pybind11)
diff --git a/test/c++/CMakeLists.txt b/test/c++/CMakeLists.txt
index 9fcbe7481..a17b85d04 100644
--- a/test/c++/CMakeLists.txt
+++ b/test/c++/CMakeLists.txt
@@ -1,15 +1,26 @@
 # Copy h5 files to binary dir
-file(GLOB_RECURSE all_h5_ref_files RELATIVE ${CMAKE_CURRENT_SOURCE_DIR} *.ref.h5)
-foreach(file ${all_h5_ref_files})
-  configure_file(${file} ${file} COPYONLY)
-endforeach()
+if(HDF5Support)
+  find_package(HDF5 REQUIRED COMPONENTS C HL)
+  file(GLOB_RECURSE all_h5_ref_files RELATIVE ${CMAKE_CURRENT_SOURCE_DIR} *.ref.h5)
+  foreach(file ${all_h5_ref_files})
+    configure_file(${file} ${file} COPYONLY)
+  endforeach()
 
+endif()
 # List of all tests
 file(GLOB_RECURSE all_tests RELATIVE ${CMAKE_CURRENT_SOURCE_DIR} *.cpp)
 if(NOT CudaSupport)
   list(FILTER all_tests EXCLUDE REGEX "nda_cu")
 endif()
 
+if(NOT MPISupport)
+  list(FILTER all_tests EXCLUDE REGEX "mpi")
+endif()
+
+if(NOT HDF5Support)
+  list(FILTER all_tests EXCLUDE REGEX "h5")
+endif()
+
 macro(SetUpAllTestWithMacroDef extension macrodef)
 foreach(test ${all_tests})
   get_filename_component(test_name ${test} NAME_WE)
@@ -17,7 +28,10 @@ foreach(test ${all_tests})
   #MESSAGE("${test_name} with option ${ARGV1}")
   get_filename_component(test_dir ${test} DIRECTORY)
   add_executable(${test_name} ${test})
-  target_link_libraries(${test_name} ${PROJECT_NAME}_c gtest_main ${PROJECT_NAME}_warnings hdf5::hdf5)
+  target_link_libraries(${test_name} ${PROJECT_NAME}_c gtest_main ${PROJECT_NAME}_warnings)
+  if (HDF5Support)
+    target_link_libraries(${test_name} hdf5::hdf5)
+  endif()
   target_compile_options(${test_name}  PRIVATE "${ARGV1}")
   set_property(TARGET ${test_name} PROPERTY RUNTIME_OUTPUT_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/${test_dir})
   add_test(NAME ${test_name} COMMAND ${test_name} WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/${test_dir})
diff --git a/test/c++/nda_h5.cpp b/test/c++/nda_h5.cpp
index 94e162278..a44323a49 100644
--- a/test/c++/nda_h5.cpp
+++ b/test/c++/nda_h5.cpp
@@ -13,6 +13,7 @@
 #include <h5/h5.hpp>
 #include <hdf5_hl.h>
 
+
 using namespace nda::clef::literals;
 using nda::ellipsis;
 using nda::range;