Sometimes. Not always directly. It’s a puzzle you’ll have to solve depending on your rig, and your animators’ tastes. The solution will not always be “place the pole vector differently”.
Sometimes you might need to layer things together. For instance, you could have 2 (or more) separate IK chains. One straight, one at an angle. And the straight one might carry some information that would drive the angled one, and you could calculate the offset between the two. Sort of like cobbling together a custom IK solver using multiple components.
I’ve done something like that to separate the bend and twist components of a robot leg in the past, where certain segments had single-axis mechanical constraints, but had to add up to acting like a regular limb.
Or sometimes it might make sense to calculate the pole vector at a 90 degree angle from the plane. For example, this is often useful for quadrupeds, or to make flipless IK. But animators often don’t like it.
Or sometimes you might need to rig straight, and then pose the rig with an offset BEFORE skinning, so your chain isn’t in neutral position. But you could store the offset, and hide it from your animators, so they are still working in neutral.
Or similarly, rig it with the bend as you have it. The pole vector will be placed inwards. Then move the parent group of the pole vector where you want it. Then calculate how much the chain rotated, and just use the twist to counter back to neutral pose. The knee won’t be pointing precisely at the pole vector controller anymore. But if it isn’t too extreme, that won’t matter. The animator just needs the relative control anyway.
(Sorry if that may have been long and confusing. But the short summary is don’t always just think about placing the pole vector. There are other components of the puzzle you can alter or combine.)