Multicore machines equipped with accelerators are becoming increasingly popular in the High Performance Computing comunity. Due to the lack of consensus regarding the definition of a standard programming model for such machines, an increasing number of HPC developers are manually combining multiple programming environments to effectively use every underlying processing unit. In this document we present a framework which is able to dynamically alocate the computing ressources to the upper layers with great flexibility, in order to allow parallel applications to be seamlessly developed by composing existing parallel kernels.