Brain-Computer Interface (BCI) systems enable direct communication between the human brain and external devices by utilizing neural signals to control machines or computers. Among the various BCI paradigms, motor imagery (MI) BCI has garnered significant attention due to its potential in real-world applications, such as assistive technologies and smart home control. However, most existing studies on consumer-grade real-time MI-BCI systems have been limited to two-class discrimination tasks (e.g., left vs. right hand movement), which constrains their applicability to more complex, real-world scenarios. In this pilot study, we explore the feasibility of a real-time MI-BCI system capable of distinguishing among four MI classes (left hand, right hand, feet, and idle) using a dry-electrode, 8-channel EEG device (Unicorn Hybrid Black). A hybrid CNN-LSTM deep learning model was employed for classification analysis. Our results revealed a modest above-chance classification performance of 40.9% for the offline session and 35.9% for the online session, with significant variability across subjects. Further analysis indicated that the strength and clarity of Event-Related Desynchronization (ERD) patterns associated with motor imagery were critical factors influencing performance. These findings suggest that while the proposed system shows significant promise, addressing the challenges of consistent and reliable performance, especially among new users, is essential for its real-world application.