The parameter ordering of a call to element_wise within one of the element_sub overloads is incorrect.

Since the function signature of element_wise is:

void element_wise(ublas::matrix<T, L1, opencl::storage> const &a,
	ublas::matrix<T, L2, opencl::storage> const &b,
	ublas::matrix<T, L3, opencl::storage> &result,
	O op, compute::command_queue& queue)

The call in the first element_sub overload should be changed to:
element_wise(a, b, result, compute::minus<T>(), queue);
as opposed to
element_wise(a, b, compute::minus<T>(), result, queue);