This paper presents a Hardware-Aware Neural Architecture Search (HW-NAS) framework for automating the design of deep learning models for texture classification using a multisensory glove. The glove is equipped with five tactile sensors and five inertial measurement units (IMUs), providing rich multimodal input. The proposed approach applies HW-NAS to both Multi-Layer Perceptrons (MLPs) and one-dimensional Convolutional Neural Networks (1D-CNNs), systematically exploring and evaluating candidate architectures. The resulting optimized models are compared to assess their efficacy in classifying textures data. Experimental results show that the optimized 1D-CNN outperforms the MLP by approximately 1.2% in classification accuracy, with only a 2.48 ms increase in latency.