This position paper discusses the challenges and opportunities of simulating kernel P systems (kP systems) using two powerful agent-based modelling frameworks on parallel architectures: FLAME (Flexible Large Scale Agent Modelling Environment) and FLAME GPU, its extension for High Performance Computing (HPC) platforms based on general purpose graphic processing units. These template-based simulation environments have been successfully used in several computationally demanding applications, ranging from macroeconomics to biological systems, thus efforts were put in translating P system models into Communicating Stream X-Machines (CSXM), the core theoretical computational devices of FLAME and FLAME GPU. Following this, translation into FLAME agents for kernel P systems having only rewriting and communication rules was proved always possible and a translation tool integrated in kPWorkbench, the software simulator for kernel P systems, was implemented. However, there are other useful features of kP systems, such as the presence of structure changing rules, like membrane division or dissolution, link creation or destruction rules, that could be taken into account, in order to exploit the full expressiveness and dynamic power of these models.