Standard real-time models do not consider the fact that a chosen technical deadline is different from the critical latency where the service utility becomes zero.  But this fact is rather important for engineering practice.
In this talk we present a tolerance-based refinement of the real-time model.  By doing so we make the process of deriving the estimation of the critical latency explicit.  The difference between the technical deadline and the critical latency is a measure for the safety margin of the system.  This safety margin is important for both, soft real-time and hard real-time systems, though with different quantities and qualities.  Furthermore, we explain why the critical latency can hardly be quantified by a concrete value.  However, we demonstrate how to derive reasonable estimates for it.  We use a concrete application to show how the distinctive knowledge of the critical latency and the technical deadline are useful for real-time scheduling.  As an outlook we argue by examples how this tolerance-based model is also applicable beyond just real-time performance, like optimised energy management.