This paper presents an efficient optimization pipeline for one-dimensional convolutional neural networks (1D-CNNs) for real-time tactile data processing. The approach employs knowledge distillation (KD), allowing a lightweight student model to learn from a high-performing teacher network, maintaining high accuracy while significantly reducing model complexity. To further enhance computational efficiency, the student network undergoes structured weight pruning and architectural reshaping. The approach was evaluated on a tactile dataset collected from a multisensory glove. When deployed on the Nano 33 BLE Sense, the optimized model achieved a 32% reduction in memory usage and a 30% improvement in inference speed, with only a minor drop in accuracy \(\approx \) 6%. These results demonstrate the potential of distilled and restructured neural networks in enabling real-time tactile perception on resource-constrained platforms.