Federated learning is an effective strategy for learning from distributed data without aggregation at a central site. Current approaches in federated learning either create a specialized application for each algorithm or use a distributed environment to share and run code among parties. The first approach is not easily generalizable and limits the possibility of end-users to experiment, while the second is vulnerable to security threats such as malicious code execution. In NextGen, we decompose algorithms into predefined tasks and share tasks rather than code. Data owners will maintain complete control over their data, including the option to determine which data to share and when.