A system that executes linear pipelines can benefit from domain-specific knowledge. This knowledge can be expressed with domain-specific directives and the design of the runtime system can exploit this knowledge to perform domain-specific optimizations. In this way, parallel programming is simplified since all low-level details are hidden from the user, and the proper configuration of the runtime system is easier since the parameters that determine the performance of linear pipelines are explicit to the user. The evaluation on a 44-core machine shows that a dynamic scheduler that is tailored to linear pipelines significantly outperforms the state-of-the-art for a set of popular fine-grained linear pipelines.