New Feature: Automatic porting of univariate special mathematical functions to the device.

I am developing a methodology that will allow automatic code generation of approximant algorithms for computing arbitrary univariate special mathematical functions on the device with high speed at the cost of a controlled loss in accuracy. 

**I wonder if this feature would be useful in a future release of OpenCL?**

It could be used to port functions one at a time or to create a library of special mathematical functions. 

The issue names no files, tests, or entry points. Start by reviewing the OpenCL-SDK structure and existing device-function support, then establish a concrete function, accuracy target, generation workflow, and acceptance criteria before implementation.

Written by the indexing model from the issue text.