Real-time safety-critical systems are currently engineered in a determ
instic and conserative fashion. Although successful in many applicatio
n areas, this approach leads to inefficient designs that are hard to m
odify. The paper proposes an alternative approach to developing such s
ystems, which gives much greater flexibility in the design process and
implementation, without losing the ability to guarantee timing proper
ties and to analyse safety properties. It is also possible to produce
implementations with known timing and safety properties. The paper ide
ntifies the research issues which must be addressed in order to realis
e systems in this fashion.