Nowadays, the amount of data increases tremendously. Extracting information and generating knowledge from this data is a great challenge. To cope with this issue – oftentimes referred to as big data problem – we need effective means for efficient data integration, data processing, and data analysis. To enable flexible, explorative and ad-hoc data processing, several data mashup approaches and tools have been developed in the past. One of these tools is FlexMash – a data mashup tool developed at the University of Stuttgart. By offering domain-specific graphical modeling as well as a pattern-based execution, FlexMash enables usage by a wide range of users, both domain experts and technical experts. The core idea of FlexMash is a flexible execution of data mashups using different, user-requirement-dependent execution components. In this paper, we present a new approach for on-demand, automated provisioning of these components in a cloud computing environment using the Topology and Orchestration Specification for Cloud Applications. This enables many advantages for mashup execution such as scalability, availability and cost savings.