Edge clouds handle data and computations closer to its source and users. Applications like industrial automation, bring new challenges and require solutions tailored for computation-centric edge cloud networks. In this paper we build on existing edge and fog computing models and develop a solution to predict and store data in edge resource caches for upcoming computations. Our solution is based on grouping caches according to the workloads they serve. We further develop methods for populating the caches and ensuring the coherence of the cached data. We evaluate the performance of our grouping mechanisms and show that they bring significant performance gains, both in terms of network traffic and access latency.