As indicated in the documentation, the main entry points for CLyther are the clyther.task and clyther.kernel - after the function is decorated with one of them, the function will be compiled in OpenCL when called.
CLyther is a compiler of a subset of the Python language. It compiles your Python subset code in OpenCL, so the actual kernel runtime will not (or should not) differ greatly between interfaces with OpenCL. The actual CLyther overhead (like all Python interfaces) comes from calling OpenCL functions or moving data between CLyther / Python and OpenCL.
Benchmarks showing CLyther performance are available in the documentation . The source archive contains the C ++ and FORTRAN versions of the test program, the solution of the Laplace equation, so you can use them to reproduce the test results yourself.
Personally, I believe that you can use CLyther effectively for most problems requiring OpenCL computing.