What is the most "elegant" way to implement a class with both CPU and GPU parallelizations in Numba?

Let's just assume that I have implemented two different versions of the same class with Numba:

class cpu_foo(object):
    def init(self, data):
        # initialize the class...

    def compute(self, parameters):
        call_to_parallel_cpu_function(parameters)

class gpu_foo(object):
    def init(self, data):
        # initialize the class...
        # ...with some fine tunings for the gpu computation

    def compute(self, parameters):
        call_to_parallel_cuda_function(parameters)

And then, depending on the machine I'm working on, I call either one version or another.

I know that it's an horrible practice to have 2 separate versions of the same code, but I'm unsure on what would be the most "standard" way to implement a class capable of:

• detect if there is a CUDA capable device
• if yes, initialize as a gpu class
• if not, initialize as a cpu class

In general, I was unable to find any design indication for any language about this topic... am I missing something?