Deep neural networks (DNNs) have gained significant attention due to their remarkable capability in learning complex features from large datasets, particularly in tasks like object classification. However, deploying these DNNs on resource-constrained edge/embedded platforms poses challenges due to their complexity and high computational demands. This research aims to address these challenges by proposing optimization techniques tailored for such platforms. Specifically, we conduct a comprehensive literature survey to identify suitable DNN architectures for embedded devices, focusing on the ResNet-18 model. We then implement adaptive filter pruning techniques to reduce model complexity while preserving critical features. Additionally, we investigate quantization techniques to further optimize model performance. Our approach is evaluated and compared against state-of-the-art pruning models, demonstrating its effectiveness in achieving efficient and accurate object classification on resource-constrained devices. Our proposed model has achieved the accuracy of 57% when the FLOPS is reduced by 60.8% in comparison with other state-of-the-art models.One of which is LCCL having accuracy of 42% with the FLOPS reduction of 35.57%.