Personally, if the class is small (for example just an helper) and is not to be used anywhere else, I would prefer doing an inner class. However, this is mostly a matter of opinion.
I think the best in these case is to make sure everyone in your dev team work the same way so it is easier for everyone to debug.
Note that there is a difference between inner class and nested class. A nested (static) class is an inner class declared static, while a simple inner class is normally not static.
Nested static class can be accessed anywhere using Class.NestedStaticClass
.
See Nested class documentation for more details and example.
Here an interesting quote from the link I gave u before :
Serialization of inner classes, including local and anonymous classes,
is strongly discouraged. When the Java compiler compiles certain
constructs, such as inner classes, it creates synthetic constructs;
these are classes, methods, fields, and other constructs that do not
have a corresponding construct in the source code. Synthetic
constructs enable Java compilers to implement new Java language
features without changes to the JVM. However, synthetic constructs can
vary among different Java compiler implementations, which means that
.class files can vary among different implementations as well.
Consequently, you may have compatibility issues if you serialize an
inner class and then deserialize it with a different JRE
implementation. See the section Implicit and Synthetic Parameters in
the section Obtaining Names of Method Parameters for more information
about the synthetic constructs generated when an inner class is
compiled.
You might also consider using Anonymous inner class. An anonymous inner class is a class coded directly in the instanciation. For example
new ParentClassName(constructorArgs) {
members..
}