If the platforms already exist, it's often the case that there may already be defined model-driven processes that can be used off the shelf. However, there will be cases in which the team has to adapt a model-driven process, for one reason or another, even while development is continuing. For example, if an architect on the project decides that a change in the target platform is necessary, then someone may have to change the bridge that creates the models for that platform. Ideally, though, the developers carrying out the model-driven process wouldn't even have to know, because they simply continue to formalize knowledge in each domain and build bridges, only applying the modified bridge for a given platform during the last step. Typically, this process will happen before the execution of the actual model-driven process.
After you've created your (customized) process, you can proceed by repeatedly applying the two high-level steps knowledge formalization and building bridges as suggested by the domain chart in Figure 11-1. As you go, you may have to adapt the process, by adapting some of the bridges, for example, or by changing platforms and their corresponding metamodels. In any case, you need to make sure that you lose as little of the formalized knowledge as possible.