If kernel was impreted as a "logical operation" there is going to be plenty of those going around with full-application implementation.

Is there some practical limit in the count of kernels loaded simultaneously in memory?

I'm asking so that I know whether the code needs to be "merged" to less kernels that contain the conditional blocks to execute proper code, or can we keep them simply separated operations each.