Studies have developed various types of soft robotic gloves for hand rehabilitation in recent years. Most soft actuators achieved a sufficient thumb flexion assist while lacking opposition support, which requires the coordination of thumb flexion and abduction-adduction. The difficulties for thumb support lie in the intrinsic complexity of thumb movements and spatial restriction of the hand. To realize multiple degrees of freedom of the thumb and make effective use of the limited space of the hand's dorsal side, we optimized and compared two approaches for thumb support. The combination approach used two independent soft actuators for thumb flexion and abduction-adduction support, respectively. The all-in-one approach used one single soft actuator to assist motions in different directions. We designed the soft actuators for each approach based on the thumb's biomechanical characteristics and evaluated their thumb flexion, abduction support performance in terms of the range of motion (RoM) and force output, and the opposition support performance using an enhanced Kapandji test. The results showed a larger abduction RoM and force output of the composition approach and a higher Kapandji score of the all-in-one approach, indicating that the two approaches might be applicable for thumb support but have the advantage in different rehabilitation stages.