We will build on KeLP's data structures for structured irregular grids. The idea is to arrange for components to build a task graph data structure at run time, analogous to KeLP's MotionPlan. The task graph will represent a large-grain dataflow graph. We will use this to increase scope for adaptive run-time scheduling, and also investigate off-line optimisation and automatic selection of placement for intermediate Regions. We will develop Imperial's run-time optimisation work. By extending it from sequential to parallel composition we will integrate task and data parallelism in a common optimisation framework which will naturally handle adaptive block-irregular applications. We will also extend the role of performance models in cross-component resource management. as developed in the P3L compiler at Pisa to deal not only with task graphs but also heterogeneous clusters of processors and irregular data.