With the rapid development of various wireless sensors and Internet-of-things devices, a massive amount of time-varying data is generated and continuously streamed to the cloud. Given the transmission latency and cost, real-time or near real-time data analytics using the conventional cloud-centric approach are challenging. Hierarchical computing models have been introduced to reduce the amount of data uploaded to the cloud and enable timely data analysis at different network endpoints. Edge computing enables analytics to be physically close to the data source at sensing and aggregation nodes. The limited computational resources on the sensor nodes and aggregation nodes motivate the development of lightweight, energy-efficient algorithms. In this thesis work, we develop a two-tier data reduction framework that exploits the spatial and temporal characteristics of the monitored environment using Deep Neural Network (DNN) architectures. Dual prediction (DP), outlier detection, and data compression (DC) techniques are adopted to reduce the number of transmissions while ensuring the quality of the collected sensor data. The dual prediction exploits the temporal correlation at the sensing nodes to minimize communication between sensor nodes and the cluster aggregation node. At the same time, outlier detection uses the spatiotemporal characteristic among multiple sensor nodes to enable real-time event detection. We further integrate data compression to reduce redundant data streams between the aggregation node and the cloud. Moreover, to accelerate the computations of the proposed DNN-based framework while considering the limited computational resources and power consumption, we introduce a novel unified CORDIC-based computing kernel. Through experiments in an FPGA-based testbed, we show that the proposed scheme could achieve up to a 70% reduction in streamed traffic. In addition, the sensor data could be recovered at the receiving network endpoint with minimal accuracy loss.