Collison detection is a wide -ranging real-world application. It is one of the key components used in gaming, simulation and animation. Efficient algorithms are required for collision detection as it is repeatedly executed throughout the course of an applic ation. Moreover, due to its computationally intensive nature researchers are investigating ways to reduce its execution time. This paper furthers those research works by devising a parallel CPU-GPU implementation of both broad and narrow phase collision detection with heterogenous workload sharing. An important aspect of co-scheduling is to determine an optimal CPU-GPU partition ratio. We also showcase a successive approximation approach for CPU-GPU implementation of collision detection. The paper demonstrates that the framework is not only applicable to CPU/GPU systems but to other system configuration obtaining a peak performance improvement in the range of 18%.