'Margin' is a term used describe distance between the line and the nearest point of the two classes. The best decision surface maximizes the margin, or distance between disparate classes.
SVM prioritizes classification first, then optimizes for margin. IOW, will get classification right first, then ensure max distance between classes.
Are often ignored to optimize for separating line being drawn in the best fit for the overall data set. SVM will somewhat mediate the effect of outliers.