Business transactions involve multiple partners coordinating and interacting with each other. These transactions have hierarchies of activities which need to be orchestrated. Usual database approaches (e.g.,checkpoint, rollback) are not applicable to handle faults in a long running transaction due to interaction with multiple partners. The compensation mechanism handles faults that can arise in a long running transaction. Based on the framework of Hoare's CSP process algebra, Butler et al introduced Compensating CSP (cCSP), a language to model long-running transactions. The language introduces a method to declare a transaction as a process and it has constructs for orchestration of compensation. Butler et al also defines a trace semantics for cCSP. In this thesis, the semantic models of compensating CSP are extended by defining an operational semantics, describing how the state of a program changes during its execution. The semantics is encoded into Prolog to animate the specification. The semantic models are further extended to define the synchronisation of processes. The notion of partial behaviour is defined to model the behaviour of deadlock that arises during process synchronisation. A correspondence relationship is then defined between the semantic models and proved by using structural induction. Proving the correspondence means that any of the presentation can be accepted as a primary definition of the meaning of the language and each definition can be used correctly at different times, and for different purposes. The semantic models and their relationships are mechanised by using the theorem prover PVS. The semantic models are embedded in PVS by using Shallow embedding. The relationships between semantic models are proved by mutual structural induction. The mechanisation overcomes the problems in hand proofs and improves the scalability of the approach.