Traditionally, missile autopilots have been designed using linear control approaches with gain scheduling. Moreover, three single axis autopilots are usually designed without considering the interaction among the motion axes. Such designs cannot handle the coupling among pitch-yaw-roll channels, especially under high angles of attack occurring in high maneuver zones. In most research studies, realistic factors like fin saturation, limitation of gimbal freedom etc are not considered. Our research work contributed a nonlinear multivariable approach to the design of an autopilot for a realistic missile that overcomes these difficulties. At first, exact input-output (IO) feedback linearization and decoupling is carried out for the dynamic IO characteristics of the inner rate loop of the pitch and yaw channels. This enables the design of scalar linear controllers for the inner rate loops. However performance deteriorates when the plant model is perturbed, due to aerodynamic uncertainties, from the nominal model. The robust IO linearization techniques are developed using H_inf and sliding mode techniques.