The kernel also has parameters that need tuning i.e. the gamma parameter. So in the bottom picture at the left, we have just a basic linear kernel with C parameter set to one. In the other two kernel they are not linear but rather RBF( Radial Basis Function) we have another parameter called gamma. So as we can see visually, with a smaller gamma value like for the one in the center there is less complexity. There are also more misclassifications allowed in this model. But with the right model which has a larger model, there is over fitting, more complexity.