Motor Imagery (MI)-based Brain-Computer Interfaces (BCIs) have garnered considerable interest for facilitating direct neurological control of external devices, especially in assistive and rehabilitative technologies. However, the proficient classification of non-stationary, low-amplitude EEG signals continue to remain a significant difficulty. This study introduces a hybrid framework that combines Filter Bank Common Spatial Pattern (FBCSP) with a 1D Convolutional Neural Network (CNN) to improve the classification of motor imagery signals. Electroencephalogram data were acquired from four participants utilizing a 14-channel Emotive EPOC headset during two-class motor imaging tasks (left versus right hand imagery). The EEG samples were bandpass filtered into three frequency sub-bands (8-12 Hz, 12-16 Hz, and 16-30 Hz), and to extract discriminative spatial features the CSP was applied to each band. These features were combined and normalized before being entered into a lightweight CNN model for classification. The model was trained with the Adam optimizer and evaluated using standard metrics. Subject-specific results showed high classification ability, with accuracy approaching 100% for some individuals and an average accuracy above 90% across most subjects. The proposed FBCSP + CNN pipeline effectively captures spatial-spectral patterns in EEG data while being computationally inexpensive, making it ideal for real-time BCI applications that use consumer-grade EEG sensors. These findings emphasize the utility of hybrid handcrafted-deep learning models in actual MI-BCI systems.