With the increasing complexity of computing tasks and the rise in data volume, current embedded and realtime systems, which are armed with multiple kinds of architectures, require parallel computing capability. However, the mainstream parallel programming models like OpenCL and MPI are too complex for these embedded and realtime systems. In this paper, we present a heterogeneous parallel programming framework Paor and implement it on RTEMS operating system. Paor provides several easy-to-use APIs to facilitate that computing devices of different architectures can collaborate in executing computing-intensive or data-intensive tasks on RTEMS operating system. Moreover, Paor provides supports for computational operators, including standard operators based on BLAS (Basic Linear Algebra Subprograms) and user-defined operators, thereby facilitating the parallelization of conventional mathematical computations. Experimental results in a heterogeneous computing environment show that Paor performs well on both computing-intensive and data-intensive tasks.