Data-centric business process models couple data and control flow to specify flexible business processes. However, it can be difficult to predict the actual behavior of a data-centric model, since the global process is typically distributed over several data elements and possibly specified in a declarative way. We therefore envision a data-centric process modeling approach in which the default behavior of the process is first specified in a classical, imperative process notation, which is then transformed to a declarative, data-centric process model that can be further refined into a complete model. To support this vision, we define a semi-automated approach to synthesize an object-centric design from a business process model that specifies the flow of multiple stateful objects between activities. The object-centric design specifies in an imperative way the life cycles of the objects and the object interactions. Next, we define a mapping from an object-centric design to a declarative Guard-Stage-Milestone schema, which can be refined into a complete specification of a data-centric BPM system. The synthesis approach has been implemented and tested using a graph transformation tool.