The scheduling of real-time systems requires knowing the Worst Case Execution Time (WCET) of their tasks. WCET analysers compute timings by analysis of the low level behaviour of the target task. This document presents improvements in the WCET computation allowed by taking into account high level behaviours of the tasks. We first classify high level knowledge according to the relevance with respect to WCET estimation. We then propose a systematic method to bring this information back to the low level on which operate most WCET analysers. This approach separates the concerns of stating properties, integrating properties and computing a WCET aware of these properties.