Towards a Domain-Specific Modelling Environment for Reinforcement Learning