This paper describes a dynamic and lightweight compiler able to guide the execution of highly dynamic parallel programs at runtime without the need for a full-fledged Just-In-Time compiler. The proposed approach allows us to apply profitable optimizations that could not be used at compile-time due to lack of sufficient information. It also enables us to optimize code fragments multiple times, depending on the specific conditions of the execution environment at runtime, while limiting the performance overhead to a negligible amount.