In this paper, we propose ProceThings, a new middleware platform to provide smart community services by utilizing computational resources of IoT devices in a target community area. To realize ProceThings, we address three key challenges: (1) dynamic load balance management among numerous IoT devices; (2) distributed task assignment/execution over the IoT devices taking into account fault-Tolerance; and (3) fulfillment of a service level agreement (SLA) for each service. For (1), ProceThings employs a heuristic monitoring mechanism which hierarchically aggregates load and resource conditions from all IoT devices in the target area (or belonging to a service). For (2), ProceThings employs a cluster-based architecture where proximity IoT devices are grouped into clusters with a fail over function, where they are allocated processing tasks from user queries. For (3), ProceThings employs demand-Aware in-situ resource provisioning which dynamically predicts and assigns a sufficient amount of computational resources within the area where the service is provided to meet the SLA while preventing over-provisioning of resources. We have implemented a prototype of ProceThings running on commodity small computers consisting of Raspberry Pis and Intel NUCs and confirmed that the above mechanisms can properly work satisfying the corresponding SLAs when running smart community services.