Accurate classification of blood cell images is pivotal for advancing medical diagnostics. Our study introduces a robust framework that harnesses advanced deep learning techniques to enhance diagnostic accuracy and enable timely interventions in healthcare. We meticulously preprocessed a large dataset sourced from the Hospital Clinic of Barcelona, ensuring its suitability for model training. Employing an ensemble of cutting-edge deep learning architectures, including Convolutional Neural Networks (CNNs), U-Net, ResNet-50, VGG-16, DenseNet-201, Supervised Contrastive Learning, MobileNet-V2, InceptionV3, Xception, and a novel hybrid InceptionV3-Xception model. Comprehensive training and evaluation were conducted across these architectures. The results demonstrated significant improvements in blood cell image classification, with the hybrid InceptionV3-Xception model achieving a remarkable test accuracy of 98.51%. Furthermore, the study includes a detailed analysis of the effects of data shuffling on model performance, offering critical insights into the robustness and generalization capabilities of the proposed framework. This research contributes substantially to the field of medical image analysis, providing scalable solutions for accurate blood cell classification and advancing the prospects of healthcare diagnostics.