Heterogeneous multi-sensor patch devices should meet the requirements of various applications, and an adhesive interposer-based patch concept with integrated microscale structures is proposed to provide comfortable pressure-based reconfiguration capability where different kinds of sensor components are easily attached or detached for application-specific purposes. Furthermore, for embedding real-time edge-computing capability into a miniaturized patch device with a conventional legacy microcontroller, a multi-sensor patch interface IC is designed to include on-chip analog feature-extraction and classification engines, supporting various sensor interfaces in environmental and healthcare application. Based on the capacitor-based analog convolutional neural network for voice activity detection [1], a two-step analog feature-extraction scheme of quantized time-domain convolutional neural network (QTD-CNN) and one-shot computing analog binarized neural network (BNN) are proposed to minimize the leakage problem of analog capacitor-based methods through 1b quantization of past data without accuracy degradation, enhancing the overall computation speed. It provides flexible time window control and analog normalization capabilities to support feature extractions and classifications in environmental and healthcare applications. The designed interface IC includes five kinds of readout front-ends for chemo-resistive sensors, electro-chemical sensors, bio-potentials (ExG), photoplethysmogram (PPG), and bio-impedance (BioZ), where every path is designed to provide both wide dynamic range (DR) and compensation capability of gain and offset. A prototype of adhesive interposer-based reconfigurable patch interface is manufactured, and its real-time heterogeneous feature extractions of gas event and arrhythmia detections are experimentally demonstrated.