Do you already have a version of the matrix multiplication kernel that works on real (non-complex) numbers? If not, this is a good place to get started:
Once you have a matrix multiplication kernel that works on real numbers, to convert to a kernel that works on complex numbers you'll need to change the inner-most loop to perform a complex multiplication and add rather than a real (non-complex) mulitplication and add. The multiplication part can be a bit tricky, but the add is pretty straightforward.
If you get this working can you post your code? My google search found several others asking similar questions, but I didn't find a working example... yet!