The presence of the RM interceptors on the respective interceptor chains alone will take care that RM protocol messages are exchanged when necessary. For example, upon intercepting the first application message on the outbound interceptor chain, the RMOutInterceptor will send a CreateSequence request and only proceed with processing the original application message after it has the CreateSequenceResponse response. Furthermore, the RM interceptors are responsible for adding the Sequence headers to the application messages and, on the destination side, extracting them from the message.
This means that no changes to application code are required to make the message exchange reliable!
You can still control sequence demarcation and other aspects of the reliable exchange through configuration however. For example, while CXF by default attempts to maximize the lifetime of a sequence, thus reducing the overhead incurred by the RM protocol messages, you can enforce the use of a separate sequence per application message by configuring the RM source's sequence termination policy (setting the maximum sequence length to 1). See the Reliable Messaging Configuration Guide for more details on configuring this and other aspects of the reliable exchange.