The behavioural synthesis procedure aims to produce optimised register-transfer level datapath descriptions from an algorithmic problem definition, normally expressed in a high-level programming language. The procedure can be partitioned into a number of subtasks linked by a serial synthesis flow. Graph theoretic algorithms can be used to provide solutions to these subtasks. Many of these techniques, however, belong to a class of algorithm for which there is no exact solution computable in polynomial time. To overcome this problem, heuristics are used to constrain the solution space. The introduction of heuristics can cause the algorithm to terminate in a local cost minimum. This thesis supports a global formulation of the behavioural synthesis problem. An algorithm which can avoid local minima, simulated annealing forms the basis of the synthesis system reported. A modular software system is presented in support of this approach. A novel data structure enables multiple degrees of optimisation freedom within the datapath solution space. Synthesis primitives, tightly coupled to a solution costing mechanism directed towards the prevalent datapath implementation technologies, form the core of the system. The software is exercised over small and large-scale synthesis benchmarks. The synthesis paradigm is extended by the provision of optimisation routines capable of supporting the generation of functional pipelines.