[API Change] Refactored OpenCL backend: Kernels are now generated...

[API Change] Refactored OpenCL backend: Kernels are now generated automatically from compiled program.

A user thus no longer needs to call prog.add_kernel("...");, as this was somewhat redundant anyway.
As a nice side effect, there should be less overhead involved, as the OpenCL backend is now called only once rather than multiple times.